BJR https://doi.org/10.1259/bjr.

20210527
© 2022 The Authors. Published by the British Institute of Radiology
Received: Accepted: Published online: under the terms of the Creative Commons Attribution-­NonCommercial
28 April 2021 28 July 2021 10 December 2021 4.0 Unported License http://creativecommons.org/licenses/by-nc/4.0/,
which permits unrestricted non-­commercial reuse, provided the original
author and source are credited.

Cite this article as:

San José Estépar R. Artificial intelligence in functional imaging of the lung. Br J Radiol (2022) 10.1259/bjr.20210527.

FUNCTIONAL IMAGING OF THE LUNG SPECIAL FEATURE:

REVIEW ARTICLE
Artificial intelligence in functional imaging of the lung
RAÚL SAN JOSÉ ESTÉPAR
Applied Chest Imaging Laboratory, Department of Radiology, Brigham and Women's Hospital, Harvard Medical School, Boston, United
States

Address correspondence to: Dr Raúl San José Estépar

E-mail: rsanjose@bwh.harvard.edu

ABSTRACT
Artificial intelligence (AI) is transforming the way we perform advanced imaging. From high-­resolution image recon-
struction to predicting functional response from clinically acquired data, AI is promising to revolutionize clinical evalua-
tion of lung performance, pushing the boundary in pulmonary functional imaging for patients suffering from respiratory
conditions. In this review, we overview the current developments and expound on some of the encouraging new fron-
tiers. We focus on the recent advances in machine learning and deep learning that enable reconstructing images,
quantitating, and predicting functional responses of the lung. Finally, we shed light on the potential opportunities and
challenges ahead in adopting AI for functional lung imaging in clinical settings.

INTRODUCTION modalities utilize the underlying physics of the image

The advent of artificial intelligence (AI) heralds a new era in
digital data analysis and empowers us to interpret complex
systems through unprecedented modeling capabilities.
This power of AI has led to an explosion of applications
across multiple disciplines including computer vision, and
more recently, health care. Clinical care stands to benefit
tremendously from AI to expose meaningful relationships
in complex data sets obtained from clinical imaging to
molecular medicine. Although AI still is a nascent field in
many health-­care domains, initial applications and proof-­
of-­concept studies have shown promising and impactful
results in diagnosing different disease conditions using
only raw data sources like diagnostic imaging.1,2 Thus,
the immense analytic capacity of AI technology based on
machine learning and deep learning will power human
decision-­making and complement human cognitive capa-
bilities. Beyond equipping physicians with new abilities,
data-­driven modeling, as opposed to just model-­ based
methods, is serving as a robust paradigm that can further
improve the current cutting-­edge algorithmic approaches
in image formation, reconstruction, and post-­processing.
The functional lung imaging community is recognizing the
transformative power of AI. The data-­driven approaches
are well-­positioned to invigorate established techniques
in this field, improving robustness and often surpassing
existing capabilities. Current functional lung imaging
properties related to different disease conditions of the
lung.3 The amount of data elements generated in functional
imaging acquisitions, such as multiple MRI snapshots
during free-­breathing acquisitions or different CT energies,
is amenable to applying data-­driven approaches to discover
novel relationships across different imaging phases, which
otherwise would be difficult to identify. Various functional
imaging modalities rely on advanced acquisitions and
post-­processing approaches, and hence AI is attractive as a
primary modeling strategy.
Although AI applications in diagnostic imaging have
increased rapidly in the last few years,4,5 its clinical appli-
cation to functional lung imaging is currently more of
an evolving opportunity than a tested reality. Farhat et
al6 recently reviewed the application of deep learning in
pulmonary medicine imaging and noticed that the use of
AI in lung imaging is mostly circumscribed to chest CT
and X-­rays (CXR). In this review, we take a comprehen-
sive look at the growing interest in applying AI technology
specifically to pulmonary functional imaging and assess
the underlying concepts of the proposed methodologies
that utilize machine- and deep learning for state-­of-­the-­art
image reconstructions, functional assessment, and func-
tional imaging synthesis. We evaluate the opportunities AI
presents and weigh in on the challenges ahead for success-
fully implementing AI in pulmonary functional imaging.
 Artificial Intelligence in Functional Imaging of the Lung BJR

Figure 1. Schematic of a CNN architecture. (A) Traditional CNN architecture is used for image classification or regression. An input
image is decomposed into multiple globally aggregated features by a final-­stage fully connected neural network. Convolutional
layers are the main component in CNNs. Additional layers include data pooling to downsample the image domain, drop-­out
for model simplification, and batch normalization. (B) U-­Net architecture is a type of fully convolutional network that is widely
employed in medical imaging applications. U-­Net contains two convolutional steps: an encoder and a decoder. The encoder
reduces the input data to a latent space, and the decoder uses this information to recreate a new image. Adapted from Chartrand
et al and Zha et al10,11 with permission. CNN, convolutional neural network.

DEEP LEARNING IN MEDICAL IMAGING
The emergence of AI as a key component in medical imaging
techniques is largely propelled by vast improvements in machine
learning, specifically, deep learning. Deep learning performs a
wide variety of challenging tasks, including classification, regres-
sion, clustering, image reconstruction artifact reduction, lesion
detection, segmentation, and registration.7 Deep learning is
an extension of artificial neural networks8 as a core building
block. Deep learning gained importance in computer vision
when neural networks outperformed other methods on several
visual recognition tasks. Deep learning in medical imaging is
primarily based on the convolutional neural network (CNN)
paradigm. LeCun9 introduced the CNNs to extend the use of
neural networks from 1D signals to multi dimensional signals
like 2D or 3D volumes and provide a powerful way to learn
representations of images and solve recognition tasks. CNNs are
constructed with units of a compact kernel of neurons that slides
across an image to produce an output image map. Neurons act
like logistic regressors that generate a response at each image
location as a weighted sum of the image intensities. The kernels
define the weight of each location, and these neural kernels are
assembled in multiple channels to create a CNN convolutional
layer. Several such layers that function differently but comple-
mentary make up the CNN (Figure 1A). Information flows in
a forward fashion, and deeper and deeper layers aggregate it in
a non-­linear manner. The success of CNNs in medical imaging
inspired the development of other deep learning paradigms to
exploit the various aspects of the information flowing through
the network. A few examples of such advanced network methods
are recurrent neural networks (RNN), autoencoders (AE), and its
variations like U-­Nets, generative adversarial networks (GANs),
and more recently, transformers, among others.12–14 Figure 1B
illustrates the architecture of a U-­Net used in medical appli-
cations to generate an output image from an input image after
aggregating information at different scales. For more informa-
tion, we refer the readers to the recent reviews of deep learning
in radiology.5,10,12,15
Machine learning approaches can be classified into four major
categories depending on the nature of the problem being solved
and the data elements used as part of the training, viz. supervised
learning, unsupervised learning, semi-­supervised learning, and
reinforcement learning. A model maps a set of inputs to given
outputs in supervised learning and requires annotated data

2 of 14 birpublications.org/bjr Br J Radiol;95:20210527
 BJR
sets. Unsupervised learning aims at finding structure in a data
set, as it is common in clustering problems. Semi-­supervised
learning has emerged as an exciting approach that combines
supervised and unsupervised techniques to take advantage of
non-­annotated datasets that can improve supervised learning
by matching specific characteristics of the non-­annotated data
set. An example of semi-­supervised learning is image-­to-­image
translation using GANs.16 Finally, reinforcement learning is
based on agents that learn from their environments through
trial and error while optimizing some objective functions. An
example of reinforming learning in medical imaging is landmark
detection methods.17
Finally, machine learning approaches define the model param-
eters using training data to solve an optimization problem. The
proper definition of the training data set in terms of character-
istics, sample size, and image conditions are key to converge to
a solution that can be generalized to other data sets beyond the
training examples. This implies that machine learning needs a
thorough validation and testing of the models using data points
that have been employed in training. Different techniques known
as cross-­validation are used to check the stability of the model
when the training data change. It is essential to understand the
conditions under which the model was derived, and the modelers
need to follow good practices and careful documentation of the
training process.18
AI IN FUNCTIONAL IMAGE RECONSTRUCTION
Magnetic resonance imaging (MRI)
MRI has been a primary modality in functional lung imaging
because of its safety characteristics and the exceptional ability
to discover functional properties.19 The early challenges due
to a lack of protons and signal inhomogeneities in the lungs
have been overcome, and now MRI can be used for static and
dynamic lung imaging.20 The arrival of ultrashort TE (UTE)
MRI with sophisticated clinical hardware has advanced lung
imaging, both at the structural and functional levels.21 From
oxygen-­enhanced and hyperpolarized gases MRI for ventilation
imaging21 to Fourier Decomposition proton MRI for ventila-
tion/perfusion (V/Q) imaging and dynamic contrast enhance-
ment (DCE) MRI for microvascular perfusion,19 MRI has
become the modality of choice to examine the complex venti-
lation and perfusion functions in different pathological condi-
tions.22 Essential to MRI pulse sequence design is the need for
short echo times and the balance between acquisition time and
signal-­to-­noise ratio (SNR) that can be achieved with parallel
imaging.23 Many of the computational approaches in MRI appli-
cations have been focused on improving optimal phase encoding
from an under sampled version of the k-­space that could reduce
the acquisition time while keeping SNR levels compatible with
image quality.24 Compressed sensing techniques were developed
two decades ago for fast MRI reconstruction, and using diffusion
MRI with hyperpolarized 129Xe.25,26 In the past few years, CNNs
and Recurrent NNs have taken a prominent role in improving
static and dynamic MRI reconstruction to learning the spatio-
temporal dependencies in heavily under sampled k-­ space
data.27–31 Duan et al32 showed improved ventilation imaging
using a coarse-­ to-­
fine neural network from under sampled
3 of 14 birpublications.org/bjr
San José Estépar
k-­space.32 Reconstruction can be achieved with higher SNR
values than compressed sensing reconstruction, paving the way
for real-­time reconstruction of contrast-­enhanced MRI of the
lung. Unlike compressed sensing, CNN reconstruction models
rely on incorporating prior information learned as part of the
training process to solve the inverse reconstruction problem.33
However, the reliance on data to define a model implies that
rigorous validation is needed.34
Another area where deep learning can impact is the inherent
need to perform motion correction in dynamic MRI acqui-
sitions. For example, Fourier Decomposition MRI for V/Q
Imaging relies on registration techniques as a critical step in their
reconstruction paradigm. Likewise, different approaches have
been proposed based on traditional functional optimization that
shows stable quality results.35 Deep learning registration offers
an alternative with low computational cost during the inference
stage once the registration model is trained.36,37 Deep learning
in MRI also has been attempted to estimate quantitative tissue
parameters using quantitative susceptibility mapping (QMS) and
MRI fingerprinting to achieve more standardized biomarkers.38
Although these techniques are yet to be applied in both preclin-
ical and clinical MRI lung imaging, deep learning could catalyze
the translation of these advanced quantitative tools.
Computed tomography (CT)
Volumetric CT (VCT) has high-­density contrast between air
and tissue and is a mainstay of clinical chest radiology. The
introduction of helical multislice CT scanning facilitated spatio-
temporal 4DCT as a tool in radiation oncology for measuring
and managing overall respiratory motion.39 Patient safety is
increased because only low dose radiation is required when
combined with advanced iterative reconstruction techniques,
and hence functional CT imaging (both 4D and dual-­energy)
is preferred for broader clinical use. Like MRI reconstruction,
new AI methods are pushing ultra-­low-­dose CT image recon-
struction to another level. Major manufacturers are intro-
ducing new deep learning schemes that show higher SNR
and contrast and improved object detectability than standard
statistical or model-­based iterative techniques.40–42 New tech-
niques under development and current iterative reconstruction
approaches capable of denoising CNNs promise to improve the
image SNR further.43 In addition to supporting low-­dose image
reconstruction, deep neural networks have also been employed
to reduce breathing artifacts and enhance image quality.44 All
these advances will make temporal ultra-­low CT a safer and
more versatile functional modality in clinical applications of
CT.
Cone-­beam CT (CBCT) system is becoming a key device in the
interventional suite due to portability and high reconstruction
quality for volumetric images. In addition, deep learning is cata-
lyzing dynamic applications with real-­time reconstruction from
sparse projection data permitting real-­time ventilation imaging
in image-­ guided radiotherapy.45,46 The combination of these
improvements can open the door for these preclinical CBCT
applications to broader adoption as a lung functional imaging
modality.47
Br J Radiol;95:20210527
 Artificial Intelligence in Functional Imaging of the Lung
Dual-­energy CT scanning (DECT) with contrast agents (iodine
or Xenon) has also enabled the assessment of regional ventila-
tion and perfusion by taking advantage of the difference in linear
attenuation coefficient at different X-­ray energies.48–52 CNNs are
being applied to improve DECT imaging fundamentals related
to material decomposition,53,54 simplify dual-­energy acquisitions
based on single-­energy material decomposition55 and combine
virtual single-­energy structural imaging from dual-­energy acqui-
sitions. The translation of these techniques can expand the role
of DECT in ventilation and perfusion imaging as dual-­energy is
more readily available.
Positron emission tomography (PET-­ CT) and single-­ photon
emission computed tomography (SPECT) have also been
employed to perform V/Q imaging to improve planar lung scin-
tigraphy56 and assess pulmonary inflammation.57 Deep learning
solutions are being developed to enhance PET reconstruction
and attenuation correction58,59 ; however, up to date, no valida-
tion studies have been performed to show the impact of AI-­en-
hanced molecular imaging in the lung. Thus, this area remains an
exciting opportunity for AI in the years to come.
AI IN FUNCTIONAL QUANTIFICATION
Automated lung segmentation in functional
modalities
For a functional imaging modality, it is important to define
the structural components of the lung, such as lung field, lobar
compartments, fissures, and the bronchovascular tree, to locate
and quantitate image-­based data. Deep learning is significantly
evolving and transforming the post-­ acquisition upstream
operations necessary to resolve the lung’s structural compo-
nents to interpret and quantify regional functional markers.
Deep learning is indeed replacing the rule-­based approaches
to segment the lung60 and the lobes with more precise and reli-
able mapping methods based on CNNs that have shown more
consistent results across modalities.6 In particular, the use of
U-­nets, a specialized neural network architecture for semantic
segmentation, has provided compelling results in multiple
medical and biomedical imaging segmentation tasks.61,62 These
new approaches to image segmentation are superior in part
due to their enhanced ability to encode shape priors of the
segmented organ based on the provided training data without
explicitly modeling the shape. One example of the application
of U-­Nets to functional modalities is the use of a 2D U-­Net to
perform volumetric lung segmentation from UTE proton MRI
in a multiplane fashion.11 Despite reduced contrast around the
lung boundaries, the lung volume estimates in a set of asthmatic
and cystic fibrotic patients closely matched the reference values
(Figure 2). One caveat for the application of deep learning is the
limited availability of training data. Recently, Guo and colleagues
showed increased robustness in UTE MRI lung segmentation by
including an adaptive k-­mean after the initial U-­net segmenta-
tion.63 Robust lung segmentation in MRI is essential for quan-
titative analysis of functional parameters and its use in clinical
studies. Similarly, a multi resolution U-­Net architecture has been
proposed for robust lobar segmentation in CT images to enable
regional quantification of dynamic CT series.64,65
4 of 14 birpublications.org/bjr
BJR
Deformable image registration (DIR)
Ventilation imaging. DIR is one of the most employed methods
to assess ventilation defects from temporal imaging modalities
like 4DCT, CBCT, and MRI. DIR-­enabled CT-­based ventilation
assessment has been successfully used in radiation oncology to
avoid damage from radiation therapy as well as performing dose–
response assessment.39 Recently, MRI-­based mechanical assess-
ment of the lung via elastic registration has also been used to
assess SSc-­related fibrosis.66 Ventilation assessment using tissue
expansion metrics based on the deformation fields generated by
DIR or the differences in tissue density between the coregistered
image pairs have shown reasonable correlation with the regional
assessment of ventilation using Xenon CT48,67 and Xenon-­MRI.68
However, variability between registration approaches has led to
a poor correlation between DIR-­based ventilation metrics and
reference modalities at the voxel level.69
Traditional DIR approaches describe the mapping of two
images via a deformable field by finding the elastic transforma-
tion parameters that minimize the difference between images
acquired at different moments during the respiratory cycle.
Deformable registration in the lung has been challenging by the
complexities of describing the transformation in a parametric
way when dealing with large displacements commonly found
in registration between images acquired between TLC and FRC
while preserving known invariants like lung mass.70 Neverthe-
less, traditional methods have partially addressed lung registra-
tion with reasonable accuracy performance, albeit with complex
methodologies that lack robustness and require long computa-
tion times due to the numerical minimization needed for each
registration instance.71
Deep learning-­based deformation image registration (DLDIR)
has emerged in the last 5 years as a new paradigm for regis-
tration. One of the main advantages of DLDIR approaches is
the explicit or implicit definition of the deformation field via a
CNN that can better capture the complexities of the deforma-
tion in a particular problem with relatively low computational
needs during the inference step. DLDIR can be classified into
supervised and unsupervised registration methods. Supervised
approaches that regress the displacement vector field between
two images using a CNN model were initially employed in
DLDIR72,73R These methods were trained with previously
aligned images using either a reference method72 or synthetic
deformations.73,74 Although these approaches improve the regis-
tration computing times from minutes to just a few seconds,
their accuracy is defined by the characteristics of the reference
approach used for learning. The reported registration errors on
reference data sets are on par with their traditional techniques
that have been extensively used in ventilation studies. Unsuper-
vised registration approaches have been explored to overcome
the limitation of using an explicit reference deformation. Among
them, unsupervised DLDIR has captured the attention in the last
few years because it needs only limited training data.75 Unsu-
pervised techniques use a mismatch metric between the moving
image and the reference image within the training data, as occurs
in a traditional registration framework. A CNN model encodes
the deformation parameters, and the optimization is done over
Br J Radiol;95:20210527
 BJR San José Estépar

Figure 2. Segmentation of the lung field on oxygen-­enhanced UTE MRI images using a multiplane (axial and coronal and final
consensus) U-­Net approach in a 37-­year-­old female with cystic fibrosis. The delineation of the lung boundaries can be achieved
despite the reduced contrast. Adapted from Zha et al11 with permission. UTE, ultrashort TE.

the parameters of the CNNs rather than the deformation param-
eters. Once the training is completed, the CNN is employed to
generate new deformation parameters from unseen data sets. De
Vos and colleagues37 pioneered this framework in lung regis-
tration using the multiscale ConvNet architecture (Figure 3). A
similar approach has been shown to be feasible to register CT to
CBCT and CBCT to CBCT76 and one-­shot methods have been
tailored to track periodic breathing motion patterns.77 Finally,
Fu et al77 proposed a LungRegtNet for 4DCT registration that
employs vascular landmarks to achieve superior performance
compared to current methods based on unsupervised registra-
tion in the DIRLAb data set.78
The new breed of lung DLDIR approaches can lead to higher
accuracy and more robust registration results that could improve
the assessment of regional ventilation at the voxel level; however,
extensive validation studies in larger prospective samples should
be conducted to confirm this possibility.69,79 Inaccurate regis-
trations can result in lung tissue being mapped to blood vessel
voxels which will cause artifacts in the CT-­ventilation image in
both the Jacobian and HU formulations. Without any doubt,
the most exciting characteristic of DLDIR is the need for low
computation to resolve a deformation field once the method has
been trained. This opens the opportunity for bringing DIR-­based
ventilation metrics closer to the patient point-­ of-­
care when
applied to lower-­cost setups like 4D CBCT. These exciting tech-
niques are potential modalities for ventilation assessment during
treatment in the near future.39
Multiparametric assessment. Registration is also a fundamental
processing component of multiparametric structural and func-
tional imaging analyses to correlate structural changes with
functional defects in lung pathophysiology.80–82 MacNeil et al83
used volume-­matched CT and hyperpolarized helium-­3 (3He)
MRI using static and diffusion-­weighted imaging to define a
multiparametric response map (mPRM). Structural changes
measured on CT were coupled with regional MRI-­based venti-
lation and microstructure based on the apparent diffusion coef-
ficient (ADC) as shown in Figure 4. mPRM metrics were able
to reveal emphysema and small airways disease not otherwise
identified with CT or MRI, reflecting the power of multimodal
approaches in disease characterization. Registration approaches

5 of 14 birpublications.org/bjr Br J Radiol;95:20210527
 Artificial Intelligence in Functional Imaging of the Lung BJR

Figure 3. An example of Unsupervised Deep Learning Deformable Image Registration from an expiratory (moving) to an inspira-
tory (fixed) CT scan. The CNN models the deformation field depicted as a warped grid. The Jacobian map estimates the volume
change and can be used to compute ventilation maps. Registration inference can be performed in a few seconds in comparison to
classical techniques enabling real-­time deployment. Adapted from Vos BD de et al37 with permission. CNN, convolutional neural
network.

that include deep learning schemes will likely translate these
upcoming multiparametric approaches to clinical applications
beyond correlative studies.
Functional prediction
Deep learning approaches have been also postulated to predict
the functional parameters from structural modalities. Westcott
and colleagues showed how textural-­based features extracted
from a volume of interest on CT scans can predict regional venti-
latory effects in subjects with COPD.84 The method was trained
with ventilation defect labels obtained from 1H and 3He MRI
using a k-­mean approach. Different classifiers were compared,
and the most relevant features were selected in a cross-­validation
experimental setup. The AUC of the best model was 82%,
with high specificity (91%) and moderate sensitivity (49%).
Ventilation-­defect percentage (VPD) predicted by the model
and the one computed using the reference MRI modality show
a strong correlation (90%); an encouraging sign of the ability of
these approaches to offer patient-­specific information on func-
tional impairment conditions. However, it is hard to ascertain

6 of 14 birpublications.org/bjr Br J Radiol;95:20210527
 BJR San José Estépar

Figure 4. Multiparametric imaging mapping from 3He MRI and CT in COPD. Functional and structural images (top) are aligned
to produce a multiparametric Response Map (bottom). DL Registration techniques can enable accurate and real-­time response
mapping assessment. Adapted from MacNeil et al83 with permission. COPD, chronic obstructive pulmonary disease.

how stable the features proposed by this study could be gener-
alized to a larger COPD population with milder disease condi-
tions because of the limited sample size used for training. Larger
sample size and reproducibility studies are needed to define the
generalization power of the proposed features.
CNNs have also been used to extract features from CT images
that can define spirometric status in smokers with and without
COPD. Gonzalez et al85 used a three-­layer feed-­forward CNN
to predict COPD functional status based on spirometry. The
correlation between FEV1 measurements and deep learning
CT-­based measurements was 73%. Tang and colleagues used a
more complex network—a residual Network with 152 layers—to
diagnose COPD from CT volumetric imaging.86 The AUC in the
testing cohort for the best model was 86%. This result was consis-
tent with the performance reported by Gonzalez and colleagues.

7 of 14 birpublications.org/bjr Br J Radiol;95:20210527
 Artificial Intelligence in Functional Imaging of the Lung BJR

These results suggest that different architectures can extract
complementary feature information from CT imaging to predict
an outcome.
In sum, the best network architecture design in terms of combi-
nations of neural layers must strike a trade-­off between model
complexity and the ability to generalize to different populations
and imaging characteristics. Meta-­learning techniques are being
actively researched and developed to improve upon the predic-
tion of single learning techniques in multiple learning episodes
that integrates different approaches.87
AI IN FUNCTIONAL ASSESSMENT
Function assessment is one of the most exciting emerging appli-
cations of AI where a direct functional response is synthesized to
mimic a target functional modality, e.g. dual energy CT pulmo-
nary perfusion, from source modalities that require simpler or
a more direct imaging reconstruction setup. These techniques
aim to resolve intrinsic relations across functional modalities or
even the resolution of functional information from structural
modalities like CT. These approaches are referred to as image-­
to-­image translation within the AI community. They are based
on an array of supervised and semi-­supervised techniques that
range from fully CNNs like convolutional generators based on
autoencoders and U-­nets61 to Generative Adversarial Networks
(GAN)13 that combine a generator and a discriminator network.
Image-­to-­image translation techniques were borne off in the
context of computer graphics applications88 and one promi-
nent application is artificial style representation from natural
images using paired (conditional) or unpaired (cycle) GANs.16,89
In paired approaches, the training is performed in a data set
containing paired instances of the target and source modality,
while unpaired approaches can use instances from the source
and the target modalities that are not matched or even belong to
the same population of subjects.
Supervised functional synthesis
One of the first demonstrations of image translation approaches
in functional lung images has been synthesizing ventilation
imaging from 4DCT without explicit use of DIR. Unfortunately,
4DCT-­derived ventilation images are sensitive to the choice of
DIR algorithm and its accuracy.90 Direct approaches can over-
come this limitation by directly learning tissue expansion char-
acteristics from multiple snapshots across a breathing cycle.
Zhong et al91 proposed a fully convolutional model composed
of seven layers without any downsampling step to preserve the
image resolution. Despite reasonable results, fully convolutional
networks are limited to local relations between the inspiratory
and expiratory images around a voxel that could lead to inconsis-
tent results if the mismatch between inspiratory and expiratory
images is significant.
To overcome some of the limitations of fully convolutional
approaches, encoder–decoder convolutional like the U-­ net
architecture have been extensively applied in image-­to-­image
reconstruction tasks. The U-­Net architecture includes multiple
convolutional steps followed by a data down-­sampling step in the
encoder phase and up-­sampling layers in the decoder phase. Also,
information from the encoding phase at a given level is trans-
ferred to the decoder phase, similar to the fully convolutional
approach. These architectures have shown promising results in
synthesizing different functional ventilation images.92,93 Capaldi
et al93 demonstrated the use of U-­nets to estimate hyperpo-
larized noble gas MRI ventilation maps from free-­breathing
proton (1H) MRI after breathing phase sorting and interpola-
tion (Figure 5). Training, validation, and testing were done in
a set of 114 subjects with different pulmonary conditions, i.e.
asthma, COPD, bronchiectasis, and NSCLC, and healthy volun-
teers. The deep learning-­based VDP estimation showed good
agreement with reference values based on hyperpolarized 3He
MRI. In a similar fashion to Zhong et al.,89 Gerard et al92 used
a multi resolution U-­net to provide a direct estimation of the

Figure 5. Deep learning ventilation MRI for the synthesis of 3He MRI ventilation imaging from free-­breathing proton (1H) MRI. (A)
Illustration of the MRI pipeline to register and sort out free-­breathing MRI images before consumption by the image-­to-­image
U-­Net network. The training was performed to predict ventilation maps from 3He MRI. (B) Comparison between the reference
ventilation maps and DL ventilation MRI synthetic imaging for subjects with different types of obstructive airway diseases. Agree-
ment in ventilation defect percentage between modalities was high with good correspondence. Adapted from Capaldi et al93 with
permission.

8 of 14 birpublications.org/bjr Br J Radiol;95:20210527
 BJR San José Estépar

Figure 6. Illustration of image-­to-­image translation techniques for synthetic ventilation and perfusion assessment based on single
energy CT. (A) Direct Jacobian ventilation map estimation using a multi resolution deep learning approach without deformation
image registration from inspiratory and expiratory CT scans. (B) Estimation of dual-­energy perfusion maps from single energy CT
angiograms to assess perfusion defects using a functional consistency CycleGAN.

ventilation response based on the deformation Jacobian without
a DIR (Figure 6A). Unlike prior approaches, this network was
trained in an extensive database of inspiratory and expiratory CT
scans from the COPDGene cohort and showed high voxel-­wide
correlations with ventilation images based on a classical mass-­
preserving DIR approach.
Like ventilation imaging, recent studies have also shown the
use of CNN approaches to estimate lung perfusion from single
energy CT scans. Ren et al94 employed an attention U-­net archi-
tecture to synthesize albumin SPECT/CT perfusion mapping
from non-­contrast CT scans to enable functional lung avoid-
ance in radiotherapy planning.94 Their proposed neural network
is superior to the traditional U-­Net architecture and is able to
identify features from the CT domain that are compatible with
perfusion defects with moderate correlation. Despite the limited
size of the training (31 subjects) and testing data (11 subjects),
these results illustrate the ability of deep learning approaches
to estimate both ventilation and perfusion functional imaging
from routine non-­contrast CT scans under a common imaging
platform.
Adversarial functional imaging
Semi-­supervised approaches based on GANs are also under
development as an improved alternative in image translation
that aims at increasing the stability of the results of multi layered
neural networks.15 For example, Nardelli et al95 illustrated the
use of a modified conditional CycleGAN to synthesize dual-­
energy-­derived iodine perfusion maps from single energy
contrast CT scans (Figure 6B). The cycleGAN leverages both
CT imaging and structural vascular information in a setup with
three encoding CNNs and three discriminators to generate the
functional output with moderate local correlations (0.52 and
0.66 in the core and peel lung regions, respectively). Although
unpaired GAN approaches are more complex and more chal-
lenging to train due to the need to find an equilibrium point
in a min-­ max optimization problem, they seem relevant to
approximate the statistical characteristics of the image that is
being estimated. Unpaired GAN approaches can be employed
with larger databases of unpaired datasets to predict the target
functional modality without the need for scanning the same
subject with both modalities as required by plain convolutional
approaches.93,94 Thus, the application of GANs presents a greater
opportunity in the context of functional imaging. GAN-­based
learning can also be applied in various domains related to image
reconstruction and preprocessing stages like super-­resolution
and multimodal registration and modality synthesis for multi-
parametric analysis.
Opportunities and challenges
AI applications in medical imaging have exploded over the past
5 years, driven by multiple factors. First, the maturity of the deep
learning approaches exploiting non-­linear relations in the data
has been instrumental. Second, advances in optimization and
regularization techniques have made it tractable to fit models
with a large number of parameters to a limited set of training
data points. Third, the availability of methods in well-­maintained
open-­source libraries has empowered a broad community with
AI techniques, including non-­experts in the field with limited
skills. Finally, specialized computing architectures based on
Graphics Processing Units (GPUs) have delivered the necessary

9 of 14 birpublications.org/bjr Br J Radiol;95:20210527
 Artificial Intelligence in Functional Imaging of the Lung
computing power to train advanced models within a reasonable
amount of time.
While AI is still an emerging discipline in functional lung
imaging, there are clear and tangible opportunities worth
mentioning:
(1) Multifunctional assessment: AI has the potential to unleash
the power of multiple functional assessments under a single
imaging platform. Currently, ventilation and perfusion
imaging require the use of different imaging contrast
agents in CT. One potential integrated solution could be
the emerging combination of 4D CBCT and simulated
dual-­energy imaging for functional imaging. The benefits
of synthetic multifunctional assessment include reduced
radiological tests that require hard-­ to-­
obtain radioactive
contrast agents, reduced radiation exposure, and improved
care delivery as imaging synthesis is performed without the
patient as part of the radiological and clinical evaluation.
However, realizing these opportunities will require an
extensive validation process to define the interval confidence
in which the synthetic images are consistent with the
underlying functional ground truth. The outcome of the
validation studies will also determine the potential of AI-­
enabled synthetic imaging for clinical adoption and whether
it could eventually be circumscribed to narrower clinical
scenarios where an initial triage based on a sub optimal
approach might be useful.
(2) Clinical translation to low footprint radiological setups:
current functional imaging relies on advanced modalities
that require specialized equipment like hyperpolarizers.
The potential use of AI-­driven image-­to-­image translation
could bring the benefit of functional information to standard
radiological imaging modalities that are available in primary
and secondary care facilities.
(3) Novel biomarkers: functional modalities provide
voluminous multiparametric data that need to be
laboriously synthesized into specific markers of disease. AI
provides an alternative computational approach to define
novel biomarkers of the disease. Supervised CNNs can be
used to extract relevant image features that are associated
with a specific outcome. Unsupervised autoencoder
techniques can also be applied for dimensionality reduction
to define novel biomarkers from multiparametric imaging
sources.
(4) Unraveling lung structure and function: the relationship
between structure and function of the lung has been well-­
described, but we are still limited in linking the structural
changes to the functional impairment and achieving a
better characterization of the disease. Studies that combine
structural and functional modalities83,96 can take advantage
of AI as an exploratory tool to gain further insight into the
structure–function relationship.
Despite the exciting and compelling preliminary evidence prom-
ising a more significant and elaborate role for AI in pulmonary
functional imaging, several challenges remain that need to be
carefully evaluated and resolved before realizing AI as a reliable
component of clinical functional lung imaging:
10 of 14 birpublications.org/bjr
BJR
(1) Validation: data-­ driven approaches require rigorous
validation studies to gauge the generality and robustness of
the methods. Until now, most of the studies that apply AI to
functional lung imaging were performed with small datasets.
Although they provide early evidence of what AI can do,
they lack the rigor needed to qualify as bonafide approaches.
Large databases on diverse populations will be required to
train and validate the techniques before translating them
into clinical use.
(2) Model transparency: one of the major criticisms of deep
learning is a lack of transparency and interpretability. In
other words, users (clinicians and researchers) should be
able to understand the “reasoning” of the AI model; why it
renders one verdict and not the other. Model developers and
data scientists must make didactic efforts to teach the users
how the models operate and decide outcomes. Transparency
is crucial to defining a modality’s operational realm and
proactively restricting deviations from the model that can
affect image quality and diagnostic interpretability.
(3) Model robustness: one collateral effect of the lack of model
transparency is model instability to adversarial attacks
(negligible input variations resulting in significant changes
of the model output) and intrinsic model biases. Adversarial
attack prevention is an oft-­discussed topic in AI and they
pose a substantial barrier to the use of AI for image synthesis
in critical applications like diagnostic imaging.97 Careful
model design and training considerations must be taken to
avoid adversarial attacks overall if models are trained with
off-­the-­shelve components.98 In a similar fashion, biases and
disparity in functional expression may be translated into AI
systems trained with imaging data in which those underlying
biases exist.99 Understanding the specific performance
characteristics of each model is crucial to move beyond the
preclinical scenario and successfully introduce it into clinical
practice.
(4) Unlocking data silos: the unresolved complexities of
functional imaging imply that the number of training
cases is limited compared to training scenarios available
for modalities like CT and CXR. Training sample size is
a key factor in deep learning that depends on the specific
characteristics of the problem begin addressed and the model
that is used. Unlocking the available data silos is paramount
for implementing new data-­driven advances in functional
lung imaging. Open data repositories and challenges like
VAMPIRE79 are crucial for developing machine learning-­
centric approaches that improve functional lung imaging
quality and performance reasonably and reproducibly. Issues
about data integrity and privacy could be overcome with
federated solutions that enable de-­centralized AI modeling
to exploit pan-­institutional datasets.100,101
CONCLUSION
AI continues to evolve rapidly and push the limits in many
spheres, and its interest in medicine is growing exponentially
in recent years, especially in the functional imaging domain.
Public and private entities recognize this as a thrust area, and
their initiatives have begun to catalyze this field.100 The pulmo-
nary functional imaging community can benefit from this
Br J Radiol;95:20210527
 BJR San José Estépar

frenetic activity in data science as novel approaches using rich
data sets are proposed to redefine disease conditions. Machine
learning models that link imaging, functional, biomarkers, and
multi omics data can advance our understanding of the complex
and intimate connection between structure and function.102
AI can also play a transformative role in adopting functional
imaging approaches to clinical settings that are now restricted
to preclinical scenarios due to their complexity. The use of func-
tional modalities in diseases like chronic obstructive pulmonary
disease (COPD), asthma, Interstital Lung Disease (ILD), or Cystic
Fibrosis (CF) can bring a new dimensionality to define relevant
markers of disease heterogeneity and progression.82,103,104 At the
same time, the application of AI is not free of limitations and
perils stemming from the experimental nature of current tech-
niques. The reliance on vast amounts of data exemplars rather
than well-­understood “fixed” models could act as a double-­edged
sword if AI is applied without careful methodological consid-
eration. This issue is even more relevant in functional imaging
scenarios where functional metrics describe subtle pathophys-
iological processes that need to be well-­understood by the AI
developers. Therefore, a multidisciplinary approach is essential
to introduce AI in functional pulmonary imaging. Successful
incorporation of AI in functional imaging holds promise to
transform the field, delivering significant benefits in the coming
years.
COMPETING INTERESTS
Dr. San José Estépar has no conflicts of interest to disclose related
to the context of this manuscript. He is the founder and co-­owner
of Quantitative Imaging Solutions, a company that provides
image-­based consulting and develops software for data sharing
and artificial intelligence applications.
FUNDING
This study is supported by funding from the National Heart,
Lung, and Blood Institute of the NIH to Dr. San José Estépar
under the award numbers R01HL149877, R01 HL116473, and
R21HL140422. He is also supported by the National Library
Medicine award R21LM013670.

REFERENCES
1. Gulshan V, Peng L, Coram M, Stumpe
MC, Wu D, Narayanaswamy A, et al.
Development and validation of a deep
learning algorithm for detection of diabetic
retinopathy in retinal fundus Photographs.
JAMA 2016; 316: 2402–10. doi: https://doi.​
org/10.1001/jama.2016.17216
2. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter
SM, Blau HM, et al. Dermatologist-­level
classification of skin cancer with deep
neural networks. Nature 2017; 542: 115–8.
doi: https://doi.org/10.1038/nature21056
3. Ohno Y, Seo JB, Parraga G, Lee KS, Gefter
WB, Fain SB, et al. Pulmonary functional
imaging: part 1-­State-­of-­the-­Art technical
and physiologic underpinnings. Radiology
2021; 299: 508–23. doi: https://doi.org/10.​
1148/radiol.2021203711
4. Hosny A, Parmar C, Quackenbush J,
Schwartz LH, Aerts HJWL. Artificial
intelligence in radiology. Nat Rev Cancer
2018; 18: 500–10. doi: https://doi.org/10.​
1038/s41568-018-0016-5
5. Savadjiev P, Chong J, Dohan A,
Vakalopoulou M, Reinhold C, Paragios N,
et al. Demystification of AI-­driven medical
image interpretation: past, present and
future. Eur Radiol 2019; 29: 1616–24. doi:
https://doi.org/10.1007/s00330-018-5674-x
6. Farhat H, Sakr GE, Kilany R. Deep learning
applications in pulmonary medical imaging:
recent updates and insights on COVID-­19.
Mach Vis Appl 2020; 31: 53. doi: https://doi.​
org/10.1007/s00138-020-01101-5
7. Hatcher WG, Yu W. A survey of deep
learning: platforms, applications and
emerging research trends. IEEE Access 2018;
6: 24411–32. doi: https://doi.org/10.1109/​
ACCESS.2018.2830661
8. McCulloch WS, Pitts W. A logical calculus
of the ideas immanent in nervous activity.
1943. Bull Math Biol 1990; 52(1-­2): 115–33.
9. LeCun Y, Bengio Y, Hinton G. Deep
learning. Nature 2015; 521: 436–44. doi:
https://doi.org/10.1038/nature14539
10. Chartrand G, Cheng PM, Vorontsov E,
Drozdzal M, Turcotte S, Pal CJ, et al.
Deep learning: a primer for radiologists.
Radiographics 2017; 37: 2113–31. doi:
https://doi.org/10.1148/rg.2017170077
11. Zha W, Fain SB, Schiebler ML, Evans
MD, Nagle SK, Liu F. Deep convolutional
neural networks with multiplane consensus
labeling for lung function quantification
using UTE proton MRI. J Magn Reson
Imaging 2019; 50: 1169–81. doi: https://doi.​
org/10.1002/jmri.26734
12. Schmidhuber J. Deep learning in neural
networks: an overview. Neural Netw 2015;
61: 85–117. doi: https://doi.org/10.1016/j.​
neunet.2014.09.003
13. Goodfellow IJ, Pouget-­Abadie J, Mirza M,
Xu B, Warde-­Farley D, Ozair S. Generative
Adversarial networks. arXiv 2014.
14. Vaswani A, Shazeer N, Parmar N, Uszkoreit
J, Jones L, Gomez AN. Attention is all you
need. Arxiv 2017.
15. Wolterink JM, Mukhopadhyay A, Leiner T,
Vogl TJ, Bucher AM, Išgum I. Generative
Adversarial networks: a primer for
radiologists. Radiographics 2021; 41: 200151.
doi: https://doi.org/10.1148/rg.2021200151
16. Isola P, Zhu J-­Y, Zhou T, Efros AA. Image-­
to-­Image translation with conditional
Adversarial networks. arXiv 2016.
17. Ghesu F-­C, Georgescu B, Zheng Y, Grbic
S, Maier A, Hornegger J, et al. Multi-­Scale
deep reinforcement learning for real-­time
3D-­Landmark detection in CT scans.
IEEE Trans Pattern Anal Mach Intell 2019;
41: 176–89. doi: https://doi.org/10.1109/​
TPAMI.2017.2782687
18. Haibe-­Kains B, Adam GA, Hosny A,
Khodakarami F, Waldron L, et al.Massive
Analysis Quality Control (MAQC) Society
Board of Directors Transparency and
reproducibility in artificial intelligence.
Nature 2020; 586: E14–16. doi: https://doi.​
org/10.1038/s41586-020-2766-y
19. Biederer J, Mirsadraee S, Beer M, Molinari
F, Hintze C, Bauman G, et al. MRI of the
lung (3/3)-­current applications and future
perspectives. Insights Imaging 2012; 3:
373–86. doi: https://doi.org/10.1007/s13244-​
011-0142-z
20. Mills GH, Wild JM, Eberle B, Van Beek EJR,
Beek E. Functional magnetic resonance
imaging of the lung. Br J Anaesth 2003; 91:
16–30. doi: https://doi.org/10.1093/bja/aeg149
21. Kruger SJ, Nagle SK, Couch MJ, Ohno Y,
Albert M, Fain SB. Functional imaging of
the lungs with gas agents. J Magn Reson
Imaging 2016; 43: 295–315. doi: https://doi.​
org/10.1002/jmri.25002

11 of 14 birpublications.org/bjr Br J Radiol;95:20210527
 Artificial Intelligence in Functional Imaging of the Lung BJR

22. Biederer J, Heussel C, Puderbach M,
Wielpuetz M. Functional magnetic
resonance imaging of the lung. Semin Resp
Crit Care 2014; 35: 74–82. doi: https://doi.​
org/10.1055/s-0033-1363453
23. Wild JM, Marshall H, Bock M, Schad LR,
Jakob PM, Puderbach M, et al. MRI of the
lung (1/3): methods. Insights Imaging 2012;
3: 345–53. doi: https://doi.org/10.1007/​
s13244-012-0176-x
24. Ouriadov A, Guo F, McCormack DG,
Parraga G. Accelerated 129Xe MRI
morphometry of terminal airspace
enlargement: feasibility in volunteers and
those with alpha‐1 antitrypsin deficiency.
Magnet Reson Med 2020; 84: 416–26.
25. Zhang H, Xie J, Xiao S, Zhao X, Zhang
M, Shi L, et al. Lung morphometry using
hyperpolarized 129Xe Xe multi-­b diffusion
MRI with compressed sensing in healthy
subjects and patients with COPD. Med Phys
2018; 45: 3097–108. doi: https://doi.org/10.​
1002/mp.12944
26. Collier GJ, Hughes PJC, Horn FC, Chan H,
Tahir B, Norquay G, et al. Single breath‐held
acquisition of coregistered 3D 129Xe lung
ventilation and anatomical proton images of
the human lung with compressed sensing.
Magnet Reson Med 2019; 82: 342–7.
27. Wang S, Su Z, Ying L, Peng X, Zhu S, Liang
F, et al. Accelerating Magnetic Resonance
Imaging via Deep Learning. 2016 Ieee 13th
Int Symposium Biomed Imaging Isbi. 2016;
2016: 514–7.
28. Schlemper J, Caballero J, Hajnal JV,
Price AN, Rueckert D. A deep cascade of
Convolutional neural networks for dynamic
Mr image reconstruction. IEEE Trans Med
Imaging 2018; 37: 491–503. doi: https://doi.​
org/10.1109/TMI.2017.2760978
29. Zhu B, Liu JZ, Cauley SF, Rosen BR, Rosen
MS. Image reconstruction by domain-­
transform manifold learning. Nature 2018;
555: 487–92. doi: https://doi.org/10.1038/​
nature25988
30. Strack R. AI transforms image
reconstruction. Nat Methods 2018; 15: 309.
doi: https://doi.org/10.1038/nmeth.4678
31. Lin DJ, Johnson PM, Knoll F, Lui YW.
Artificial intelligence for MR image
reconstruction: an overview for clinicians. J
Magn Reson Imaging 2021; 53: 1015–28. doi:
https://doi.org/10.1002/jmri.27078
32. Duan C, Deng H, Xiao S, Xie J, Li H, Sun
X, et al. Fast and accurate reconstruction of
human lung gas MRI with deep learning.
Magn Reson Med 2019; 82: 2273–85. doi:
https://doi.org/10.1002/mrm.27889
33. Wu Y, Ma Y, Capaldi DP, Liu J, Zhao W,
Du J, et al. Incorporating prior knowledge
via volumetric deep residual network to
optimize the reconstruction of sparsely
sampled MRI. Magn Reson Imaging 2020;
66: 93–103. doi: https://doi.org/10.1016/j.​
mri.2019.03.012
34. Antun V, Renna F, Poon C, Adcock B,
Hansen AC. On instabilities of deep
learning in image reconstruction and the
potential costs of AI. Proc Natl Acad Sci U S
A 2020; 117: 30088–95. doi: https://doi.org/​
10.1073/pnas.1907377117
35. Ljimani A, Hojdis M, Stabinska J, Valentin
B, Frenken M, Appel E, et al. Analysis of
different image-­registration algorithms for
Fourier decomposition MRI in functional
lung imaging. Acta Radiol 2021; 62:
028418512094490. doi: https://doi.org/10.​
1177/0284185120944902
36. Yang X, Kwitt R, Styner M, Niethammer
M. Quicksilver: Fast predictive image
registration - A deep learning approach.
Neuroimage 2017; 158: 378–96. doi: https://​
doi.org/10.1016/j.neuroimage.2017.07.008
37. de Vos BD, Berendsen FF, Viergever MA,
Sokooti H, Staring M, Išgum I. A deep
learning framework for unsupervised affine
and deformable image registration. Med
Image Anal 2019; 52: 128–43. doi: https://​
doi.org/10.1016/j.media.2018.11.010
38. Lundervold AS, Lundervold A. An overview
of deep learning in medical imaging
focusing on MRI. Z Med Phys 2019; 29:
102–27. doi: https://doi.org/10.1016/j.​
zemedi.2018.11.002
39. Vinogradskiy Y. Ct-­Based ventilation
imaging in radiation oncology. BJR Open
2019; 1: 20180035. doi: https://doi.org/10.​
1259/bjro.20180035
40. Park C, Choo KS, Jung Y, Jeong HS, Hwang
J-­Y, Yun MS. Ct iterative vs deep learning
reconstruction: comparison of noise and
sharpness. Eur Radiol 2021; 31: 3156–64.
doi: https://doi.org/10.1007/s00330-020-​
07358-8
41. Lenfant M, Chevallier O, Comby P-­O,
Secco G, Haioun K, Ricolfi F, et al. Deep
learning versus iterative reconstruction
for CT pulmonary angiography in the
emergency setting: improved image quality
and reduced radiation dose. Diagnostics
2020; 10: 558. doi: https://doi.org/10.3390/​
diagnostics10080558
42. Brady SL, Trout AT, Somasundaram E,
Anton CG, Li Y, Dillman JR. Improving
image quality and reducing radiation dose
for pediatric CT by using deep learning
reconstruction. Radiology 2021; 298:
180–8. doi: https://doi.org/10.1148/radiol.​
2020202317
43. Hata A, Yanagawa M, Yoshida Y, Miyata T,
Tsubamoto M, Honda O, et al. Combination
of deep Learning–Based denoising and
iterative reconstruction for Ultra-­Low-­
Dose CT of the chest: image quality and
Lung-­RADS evaluation. American Journal
of Roentgenology 2020; 215: 1321–8. doi:
https://doi.org/10.2214/AJR.19.22680
44. Mori S, Hirai R, Sakata Y. Using a deep
neural network for four-­dimensional
CT artifact reduction in image-­guided
radiotherapy. Phys Med 2019; 65: 67–75. doi:
https://doi.org/10.1016/j.ejmp.2019.08.008
45. Chen G, Zhao Y, Huang Q, Gao H. 4D-­
AirNet: a temporally-­resolved CBCT
slice reconstruction method synergizing
analytical and iterative method with deep
learning. Phys Med Biol 2020; 65: 175020.
doi: https://doi.org/10.1088/1361-6560/​
ab9f60
46. Jiang Z, Chen Y, Zhang Y, Ge Y, Yin F-­F, Ren
L. Augmentation of CBCT reconstructed
from Under-­Sampled projections using deep
learning. IEEE Trans Med Imaging 2019; 38:
2705–15. doi: https://doi.org/10.1109/TMI.​
2019.2912791
47. Murrie RP, Werdiger F, Donnelley M, Lin
Y-­wei, Carnibella RP, Samarage CR, et al.
Real-­Time in vivo imaging of regional lung
function in a mouse model of cystic fibrosis
on a laboratory X-­ray source. Sci Rep 2020;
10: 447. doi: https://doi.org/10.1038/s41598-​
019-57376-w
48. Kong X, Sheng HX, Lu GM, Meinel FG,
Dyer KT, Schoepf UJ, et al. Xenon-­enhanced
dual-­energy CT lung ventilation imaging:
techniques and clinical applications. AJR
Am J Roentgenol 2014; 202: 309–17. doi:
https://doi.org/10.2214/AJR.13.11191
49. Hoey ETD, Mirsadraee S, Pepke-­Zaba J,
Jenkins DP, Gopalan D, Screaton NJ. Dual-­
Energy CT angiography for assessment of
regional pulmonary perfusion in patients
with chronic thromboembolic pulmonary
hypertension: initial experience. AJR Am J
Roentgenol 2011; 196: 524–32. doi: https://​
doi.org/10.2214/AJR.10.4842
50. Goo HW, Goo JM. Dual-­Energy CT: new
horizon in medical imaging. Korean J Radiol
2017; 18: 555–69. doi: https://doi.org/10.​
3348/kjr.2017.18.4.555
51. Fuld MK, Halaweish AF, Haynes SE, Divekar
AA, Guo J, Hoffman EA. Pulmonary
perfused blood volume with dual-­energy
CT as surrogate for pulmonary perfusion
assessed with dynamic multidetector CT.
Radiology 2013; 267: 747–56. doi: https://​
doi.org/10.1148/radiol.12112789
52. Iyer KS, Newell JD, Jin D, Fuld MK, Saha
PK, Hansdottir S, et al. Quantitative dual-­
energy computed tomography supports
a vascular etiology of smoking-­induced
inflammatory lung disease. Am J Respir
Crit Care Med 2016; 193: 652–61. doi:

12 of 14 birpublications.org/bjr Br J Radiol;95:20210527
 BJR
https://doi.org/10.1164/rccm.201506-​
1196OC
53. Zhang W, Zhang H, Wang L, Wang X, Hu
X, Cai A, et al. Image domain dual material
decomposition for dual-­energy CT using
butterfly network. Med Phys 2019; 46:
2037–51. doi: https://doi.org/10.1002/mp.​
13489
54. Abascal JFPJ, Ducros N, Pronina V, Rit S,
Rodesch P-­A, Broussaud T, et al. Material
decomposition in spectral CT using deep
learning: a Sim2Real transfer approach.
IEEE Access 2021; 9: 25632–47. doi:
https://doi.org/10.1109/ACCESS.2021.​
3056150
55. Lyu T, Zhao W, Zhu Y, Wu Z, Zhang Y, Chen
Y, et al. Estimating dual-­energy CT imaging
from single-­energy CT data with material
decomposition convolutional neural
network. Med Image Anal 2021; 70: 102001.
doi: https://doi.org/10.1016/j.media.2021.​
102001
56. Dimastromatteo J, Charles EJ, Laubach VE.
Molecular imaging of pulmonary diseases.
Respir Res 2018; 19: 17. doi: https://doi.org/​
10.1186/s12931-018-0716-0
57. Vass L, Fisk M, Lee S, Wilson FJ, Cheriyan
J, Wilkinson I. Advances in PET to assess
pulmonary inflammation: a systematic
review. Eur J Radiol 2020; 130: 109182. doi:
https://doi.org/10.1016/j.ejrad.2020.109182
58. Dong X, Lei Y, Wang T, Higgins K, Liu
T, Curran WJ, et al. Deep learning-­based
attenuation correction in the absence of
structural information for whole-­body
positron emission tomography imaging.
Phys Med Biol 2020; 65: 055011. doi: https://​
doi.org/10.1088/1361-6560/ab652c
59. Gong K, Guan J, Kim K, Zhang X, Yang J,
Seo Y. Iterative PET image reconstruction
using Convolutional neural network
representation. IEEE Trans Med Imaging
2019; 38: 675–85. doi: https://doi.org/10.​
1109/TMI.2018.2869871
60. Sluimer I, Schilham A, Prokop M,
van Ginneken B, van GB. Computer analysis
of computed tomography scans of the lung:
a survey. IEEE Trans Med Imaging 2006; 25:
385–405. doi: https://doi.org/10.1109/TMI.​
2005.862753
61. Ronneberger O, Fischer P. Brox T. U-­Net:
Convolutional Networks for Biomedical
Image Segmentation. In: Navab N, Joachim
H, Wells William M, Frangi A. F, eds.; 2015.
pp.. 234–41.
62. Isensee F, Jaeger PF, Kohl SAA, Petersen
J, Maier-­Hein KH. nnU-­Net: a self-­
configuring method for deep learning-­
based biomedical image segmentation. Nat
Methods 2021; 18: 203–11. doi: https://doi.​
org/10.1038/s41592-020-01008-z
13 of 14 birpublications.org/bjr
63. Guo F, Capaldi DP, McCormack DG,
Fenster A, Parraga G. Ultra-­Short echo-­
time magnetic resonance imaging lung
segmentation with under-­Annotations and
domain shift. Med Image Anal 2021; 72:
102107. doi: https://doi.org/10.1016/j.media.​
2021.102107
64. Gerard SE, Patton TJ, Christensen GE,
Bayouth JE, Reinhardt JM. FissureNet: a
deep learning approach for pulmonary
fissure detection in CT images. IEEE Trans
Med Imaging 2019; 38: 156–66. doi: https://​
doi.org/10.1109/TMI.2018.2858202
65. Gerard SE, Herrmann J, Kaczka DW, Musch
G, Fernandez-­Bustamante A, Reinhardt
JM. Multi-­resolution convolutional neural
networks for fully automated segmentation
of acutely injured lungs in multiple species.
Med Image Anal 2020; 60: 101592. doi:
https://doi.org/10.1016/j.media.2019.101592
66. Chassagnon G, Martin C, Marini R,
Vakalopolou M, Régent A, Mouthon L, et al.
Use of elastic registration in pulmonary
MRI for the assessment of pulmonary
fibrosis in patients with systemic sclerosis.
Radiology 2019; 291: 487–92. doi: https://​
doi.org/10.1148/radiol.2019182099
67. Reinhardt JM, Ding K, Cao K, Christensen
GE, Hoffman EA, Bodas SV. Registration-­
based estimates of local lung tissue
expansion compared to xenon CT measures
of specific ventilation. Med Image Anal 2008;
12: 752–63. doi: https://doi.org/10.1016/j.​
media.2008.03.007
68. Tahir BA, Marshall H, Hughes PJC,
Brightling CE, Collier G, Ireland RH, et al.
Comparison of CT ventilation imaging and
hyperpolarised gas MRI: effects of breathing
manoeuvre. Phys Med Biol 2019; 64: 055013.
doi: https://doi.org/10.1088/1361-6560/​
ab0145
69. Hegi-­Johnson F, de Ruysscher D, Keall P,
Hendriks L, Vinogradskiy Y, Yamamoto
T, et al. Imaging of regional ventilation:
is CT ventilation imaging the answer? A
systematic review of the validation data.
Radiother Oncol 2019; 137: 175–85. doi:
https://doi.org/10.1016/j.radonc.2019.03.010
70. Yin Y, Hoffman EA, Lin C-­L. Mass
preserving nonrigid registration of CT lung
images using cubic B-­spline. Med Phys 2009;
36: 4213–22. doi: https://doi.org/10.1118/1.​
3193526
71. Murphy K, van Ginneken B, Reinhardt JM,
Kabus S, Ding K, Deng X, et al. Evaluation
of registration methods on thoracic CT:
the EMPIRE10 challenge. IEEE Trans Med
Imaging 2011; 30: 1901–20. doi: https://doi.​
org/10.1109/TMI.2011.2158349
72. Onieva JO, Marti-­Fuster B. Image analysis
for moving organ, breast, and thoracic
San José Estépar
images, third International workshop,
RAMBO 2018, fourth International
workshop, BIA 2018, and first International
workshop, TIA 2018, held in conjunction
with MICCAI 2018, Granada, Spain,
September 16 and 20, 2018, proceedings.
Lect Notes Comput Sc 2018; 11040: 284–94.
doi: https://doi.org/10.1007/978-3-030-​
00946-5
73. Sokooti H, de VB, Berendsen F, Lelieveldt
BPF, Išgum I, Staring M. Medical image
computing and computer assisted
intervention − MICCAI 2017: 20th
International Conference, Quebec City,
Qc, Canada, September 11-­13, 2017,
proceedings, part I. Lect Notes Comput Sc
2017; 3: 232–9. doi: https://doi.org/10.1007/​
978-3-319-66182-7_27
74. Eppenhof KAJ, Pluim JPW. Pulmonary CT
registration through supervised learning
with Convolutional neural networks. IEEE
Trans Med Imaging 2019; 38: 1097–105. doi:
https://doi.org/10.1109/TMI.2018.2878316
75. Haskins G, Kruger U, Yan P. Deep learning
in medical image registration: a survey.
Mach Vis Appl 2020; 31(1-­2): 8. doi: https://​
doi.org/10.1007/s00138-020-01060-x
76. Jiang Z, Yin F-­F, Ge Y, Ren L. A multi-­scale
framework with unsupervised joint training
of convolutional neural networks for
pulmonary deformable image registration.
Phys Med Biol 2020; 65: 015011. doi: https://​
doi.org/10.1088/1361-6560/ab5da0
77. Fechter T, Baltas D. One-­Shot learning for
deformable medical image registration and
periodic motion tracking. IEEE Trans Med
Imaging 2020; 39: 2506–17. doi: https://doi.​
org/10.1109/TMI.2020.2972616
78. Castillo R, Castillo E, Fuentes D, Ahmad M,
Wood AM, Ludwig MS, et al. A reference
dataset for deformable image registration
spatial accuracy evaluation using the
COPDgene study archive. Phys Med Biol
2013; 58: 2861–77. doi: https://doi.org/10.​
1088/0031-9155/58/9/2861
79. Kipritidis J, Tahir BA, Cazoulat G, Hofman
MS, Siva S, Callahan J, et al. The vampire
challenge: a multi-­institutional validation
study of CT ventilation imaging. Med Phys
2019; 46: 1198–217. doi: https://doi.org/10.​
1002/mp.13346
80. Guo F, Svenningsen S, Kirby M, Capaldi
DP, Sheikh K, Fenster A, et al. Thoracic CT-­
MRI coregistration for regional pulmonary
structure-­function measurements of
obstructive lung disease. Med Phys 2017; 44:
1718–33. doi: https://doi.org/10.1002/mp.​
12160
81. Hoffman EA, Lynch DA, Barr RG, van Beek
EJR, Parraga G. IWPFI Investigators
Pulmonary CT and MRI phenotypes that
Br J Radiol;95:20210527
 Artificial Intelligence in Functional Imaging of the Lung
help explain chronic pulmonary obstruction
disease pathophysiology and outcomes. J
Magn Reson Imaging 2016; 43: 544–57. doi: 90.
https://doi.org/10.1002/jmri.25010
82. Washko GR, Parraga G. COPD biomarkers
and phenotypes: opportunities for better
outcomes with precision imaging. Eur Respir
J 2018; 52: 52. doi: https://doi.org/10.1183/​
13993003.01570-2018
83. MacNeil JL, Capaldi DPI, Westcott AR, 91.
Eddy RL, Barker AL, McCormack DG, et al.
Pulmonary imaging phenotypes of chronic
obstructive pulmonary disease using
multiparametric response maps. Radiology
2020; 295: 227–36. doi: https://doi.org/10.​
1148/radiol.2020191735 92.
84. Westcott A, Capaldi DPI, McCormack DG,
Ward AD, Fenster A, Parraga G. Chronic
obstructive pulmonary disease: thoracic CT
texture analysis and machine learning to
predict pulmonary ventilation. Radiology
2019; 293: 676–84. doi: https://doi.org/10.​ 93.
1148/radiol.2019190450
85. Serrano GG, Ash SY, Vegas-­Sánchez-­Ferrero
G, Onieva JO, Rahaghi FN, Ross JC.
Disease staging and prognosis in smokers
using deep learning in chest computed
tomography. Am J Respir Crit Care Med 94.
2017; 197: 193–203. doi: https://doi.org/10.​
1164/rccm.201705-0860OC
86. Tang LYW, Coxson HO, Lam S, Leipsic
J, Tam RC, Sin DD. Towards large-­scale
case-­finding: training and validation of
residual networks for detection of chronic
obstructive pulmonary disease using 95.
low-­dose CT. Lancet Digit Health 2020; 2:
e259–67. doi: https://doi.org/10.1016/S2589-​
7500(20)30064-9
87. Hospedales T, Antoniou A, Micaelli
P, Storkey A. Meta-­Learning in neural
networks: a survey. Arxiv 2020. 96.
88. Ruan C, Yuan L, Hu H, Chen D. Image
translation with dual‐directional generative
adversarial networks. IET comput. vis 2021;
15: 73–83. doi: https://doi.org/10.1049/cvi2.​
12011 97.
89. Zhu J-­Y, Park T, Isola P, Efros AA. Unpaired
Image-­to-­Image translation using Cycle-­
14 of 14 birpublications.org/bjr
Consistent Adversarial networks. ​arXiv.​org
2017;: cs.CV:.
Latifi K, Forster KM, Hoffe SE, Dilling TJ,
van Elmpt W, Dekker A, et al. Dependence
of ventilation image derived from 4D CT
on deformable image registration and
ventilation algorithms. J Appl Clin Med Phys
2013; 14: 150–62. doi: https://doi.org/10.​
1120/jacmp.v14i4.4247
Zhong Y, Vinogradskiy Y, Chen L, Myziuk
N, Castillo R, Castillo E, et al. Technical
note: deriving ventilation imaging from
4DCT by deep convolutional neural
network. Med Phys 2019; 46: 2323–9. doi:
https://doi.org/10.1002/mp.13421
Gerard SE, Reinhardt JM, Christensen
GE, Estépar RSJ. Estimating local
tissue expansion in thoracic computed
tomography images using Convolutional
neural networks. 2020; 1856–60. doi: https://​
doi.org/10.1109/ISBI45749.2020.9098413
Capaldi DPI, Guo F, Xing L, Parraga G.
Pulmonary ventilation maps generated
with Free-­breathing proton MRI and a deep
Convolutional neural network. Radiology
2021; 298: 427–38. doi: https://doi.org/10.​
1148/radiol.2020202861
Ren G, Lam S-­K, Zhang J, Xiao H, Cheung
AL-­Y, Ho W-­Y, et al. Investigation of a
novel deep Learning-­Based computed
tomography perfusion mapping framework
for functional lung avoidance radiotherapy.
Front Oncol 2021; 11: 644703. doi: https://​
doi.org/10.3389/fonc.2021.644703
Nardelli P, Estépar RSJ, Rahaghi FN, Estépar
RSJ. Functional-­Consistent CycleGAN for
CT to iodine perfusion map translation.
Lecture Notes In Computer Science 2020;
12502: 109–17. doi: https://doi.org/10.1007/​
978-3-030-62469-9_10
Eddy RL, Parraga G. Pulmonary xenon-­129
MRI: new opportunities to unravel enigmas
in respiratory medicine. Eur Respir J 2020;
55: 1901987. doi: https://doi.org/10.1183/​
13993003.01987-2019
Ma X, Niu Y, Gu L, Wang Y, Zhao Y, Bailey
J, et al. Understanding adversarial attacks on
deep learning based medical image analysis
BJR
systems. Pattern Recognit 2021; 110: 107332.
doi: https://doi.org/10.1016/j.patcog.2020.​
107332
98. Bortsova G, González-­Gonzalo C, Wetstein
SC, Dubost F, Katramados I, Hogeweg L,
et al. Adversarial attack vulnerability of
medical image analysis systems: unexplored
factors. Med Image Anal 2021; 73: 102141.
doi: https://doi.org/10.1016/j.media.2021.​
102141
99. Larrazabal AJ, Nieto N, Peterson V, Milone
DH, Ferrante E. Gender imbalance in
medical imaging datasets produces biased
classifiers for computer-­aided diagnosis.
Proc Natl Acad Sci U S A 2020; 117:
201919012. doi: https://doi.org/10.1073/​
pnas.1919012117
100. Langlotz CP, Allen B, Erickson BJ,
Kalpathy-­Cramer J, Bigelow K, Cook TS,
et al. A roadmap for foundational research
on artificial intelligence in medical imaging:
from the 2018 NIH/RSNA/ACR/The
Academy workshop. Radiology 2019; 291:
190613. doi: https://doi.org/10.1148/radiol.​
2019190613
101. Kaissis GA, Makowski MR, Rückert D,
Braren RF. Secure, privacy-­preserving and
federated machine learning in medical
imaging. Nat Mach Intell 2020; 2: 305–11.
doi: https://doi.org/10.1038/s42256-020-​
0186-1
102. Weibel ER. It takes more than cells to make
a good lung. Am J Respir Crit Care Med
2013; 187: 342–6. doi: https://doi.org/10.​
1164/rccm.201212-2260OE
103. Hatabu H, Ohno Y, Gefter WB, Parraga
G, Madore B, Lee KS, et al. Expanding
applications of pulmonary MRI in the
clinical evaluation of lung disorders:
Fleischner Society position paper. Radiology
2020; 297: 286–301. doi: https://doi.org/10.​
1148/radiol.2020201138
104. Woods JC, Wild JM, Wielpütz MO, Clancy
JP, Hatabu H, Kauczor H-­U, et al. Current
state of the art MRI for the longitudinal
assessment of cystic fibrosis. J Magn Reson
Imaging 2020; 52: 1306–20. doi: https://doi.​
org/10.1002/jmri.27030
Br J Radiol;95:20210527